RLDynamicNav

Video Link: https://youtu.be/wL3-diEXMes

Environment

• tensorflow-gpu >= 2.4
• Cython
• ORCA library, https://github.com/sybrenstuvel/Python-RVO2

Compile the LiDAR scan library

in folder DRL_dynamic_navigation/C_library, python setup.py build_ext --inplace

select algorithm

• in file DRL_dynamic_navigation/train.py revise this line: parser.add_argument('--policy', type=str, default='lidar_dqn')
• lidar_dqn: original method with DQN RL
• lidar_dqn_simple: without CNN
• sac: replace DQN with SAC
• sac_simple: SAC without CNN
• ddpg: DDPG without CNN

select reward type

parser.add_argument('--reward_simple', default=False, action='store_true')

• False, original reward
• True, without distance reward

select environment type

parser.add_argument('--complex_environment', default=True, action='store_true')

• False, original environment only with dynamic obstacles
• True, with static obstacles

SARL, RGL, CADRL

https://github.com/zw199502/navigation_among_pedestrians